AI Science Tracker
EN

Video model reproduces pressure estimates but remains limited against catheter measurements

University of California, San Francisco; California Institute of Technology; Montreal Heart Institute · 28 Aug 2024 · medicine

Discovery

Video model reproduces pressure estimates but remains limited against catheter measurements

Source check: Source checked. Not independent scientific confirmation.

Finding: A SlowFast video classifier achieved weighted AUROC 0.854 against cardiologist right-atrial-pressure estimates in 2,618 external ultrasound studies. Against catheter measurements in 319 paired studies, categorical accuracy was 49.5%, compared with 48.9% for cardiologists. This retrospective agreement study does not establish equivalence, clinical benefit or reliable invasive-pressure replacement.

Technology: SlowFast R50 convolutional video classifier; learned ultrasound-view classifiers

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Learned view classifiers select inferior-vena-cava inspiration videos; a trained SlowFast R50 model assigns three pressure categories from their static and temporal features.
Test and result
15,828 archived studies from 11,869 patients were divided by patient into 12,664 training, 1,582 validation and 1,582 test studies. The full test AUROC was 0.823; the abstract's 0.844 refers to a quality-filtered 554-study subset. External evaluation retained 2,618 Montreal studies and yielded weighted AUROC 0.854 against cardiologist categories. All 319 catheter-paired studies were kept in testing; model accuracy against binned catheter values was 49.5%, versus 48.9% for cardiologist estimates.
Limitation / still unknown
Catheter and ultrasound measurements could be up to 30 days apart. A nonsignificant difference is not an equivalence test. External reference labels were cardiologist estimates rather than catheter measurements, and external image-quality filtering excludes many original studies. Weighted metrics use UCSF class frequencies. Automated scan selection can pick a different video from the one used clinically. No prospective workflow benefit, prevention effect or safe replacement of invasive measurement is established.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://doi.org/10.1016/j.jacadv.2024.101192 · https://pmc.ncbi.nlm.nih.gov/articles/PMC11450965/

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Actual primary clinical data selection, patient split, SlowFast model Methods, Tables 1-4 and limitations read; current primary PubMed date and notice metadata checked..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse medicine discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.